People — when a departing member is the only admin of shared brains, organization admins can assume ownership or delete the blocking brains as part of removing the member.
Personal access keys — read your own captured sessions and messages with a memory:read key, or delete your own sessions with a memory:write key using the conversation or harness session ID.
Improvements9
GitHub settings — repository access, code review, and session insights are grouped under each repository's settings, with their teamspace or organization scope shown together.
Session activity — Tool mix breaks terminal usage down into the commands your agent ran. Missing inventory is shown as unavailable instead of suggesting that no tools were used. Activity also shows average agent time per turn when measured.
Session transcripts — Codex calls that run several tools now show each tool and its input separately, making the recorded work easier to inspect.
Rulebook — each rulebook header shows who it applies to, including whether it reaches you, with labels that fit better on narrow screens.
Rule suggestions — /memhub:rules-from-sessions reuses previously processed session work and handles batches more efficiently, reducing the wait for suggestions.
Claude Code plugin — the marketplace release now includes file-read rules and capture of markdown deliverables written through Bash, previously available in newer Codex/Cursor releases. Update to v0.56.1 to get these changes.
Codex plugin — session-start context now includes team rules, brain briefs, upgrade notices, and capture health before the first tool call. Available in plugin v0.58.0 and later for Codex; the Claude marketplace remains on v0.56.1.
Connected sources — connector logos are clearer, with dark-mode variants that remain visible on dark backgrounds.
Brains — empty folders now explain when they are empty in the selected teamspace, and Uncategorized distinguishes having no unfiled brains from having no brains at all.
Fixes9
PR linking — linking stays in the same review sheet, preserves your filters and scroll position, and lists each logical session once. Session previews stay above the sheet and within the screen.
Sessions Summary — personal and routed copies of the same session no longer inflate activity counts.
Session metrics — token totals count each Claude Code response once, even when its transcript splits it into several blocks. This corrects inflated totals in session activity, Insights, PR Summary, and token filters, including already captured sessions.
Routines — connecting to tools retries temporary discovery failures and refreshes expiring credentials, reducing failures before a routine starts its work.
Capture — Cursor preserves usage observations while uploads are busy, and Codex capture handles paginated and continued histories without losing or counting inherited work twice.
Artifacts — outlines omit repeated titles and abstracts so the useful sections are easier to find.
Sign-in — denied sign-in attempts explain what needs attention. If account access is refused after sign-in, Studio, onboarding, and the developer portal now explain whether to sign in again, verify your email, or use your original sign-in method, instead of leaving you on a broken screen or endless spinner.
Billing — memory imports and capture processing now respect the monthly compute limit. Work already accepted is held for retry when capacity becomes available instead of being discarded.
Payment updates — fixed a processing error that could prevent Stripe payments and subscription changes from updating billing status.
Security4
Sign-in — email verification and account-linking checks strengthen protection when signing in with different identity providers.
Sessions — sharing a brain no longer grants access to its contributors' session transcripts. Sessions are managed in Sessions rather than listed as brain content; session-derived memory remains in the brain.
Connected sources — organization connectors cannot be granted to Personal Workspaces, including when restoring a disconnected source.
Rulebook — rule outcomes stay scoped to the person and worktree that produced them, preventing another session's activity from changing the result.